Indoor 3D Reconstruction of Buildings via Azure Kinect RGB-D Camera

Abstract

With the development of 3D vision techniques, RGB-D cameras are increasingly used to allow easier and cheaper access to the third dimension. In this paper, we focus on testing the potential of the Kinect Azure RGB-D camera in the 3D reconstruction of indoor scenes. First, a series of investigations of the hardware was performed to evaluate its accuracy and precision. The results show that the measurements made with the Azure could be exploited for close-range survey applications. Second, we performed a methodological workflow for indoor reconstruction based on the Open3D framework, which was applied to two different indoor scenes. Based on the results, we can state that the quality of 3D reconstruction significantly depends on the architecture of the captured scene. This was supported by a comparison of the point cloud from the Kinect Azure with that from a terrestrial laser scanner and another from a mobile laser scanner. The results show that the average differences do not exceed 8 mm, which confirms that the Kinect Azure can be considered a 3D measurement system at least as reliable as a mobile laser scanner.

Keywords: 3D indoor reconstruction; Azure Kinect; MLS; RGB-D; TLS.

Figure 1

Overview of the general workflow.

Figure 2

Indoor office scene used for testing image averaging.

Figure 3

Operating modus for testing the influence of different ranges on measurement accuracy.

Figure 4

Design and parameters of the ChArUco board OpenCV.

Figure 5

Workflow of geometric calibration on OpenCV.

Figure 6

Diagram of the reconstruction of a 3D point clouds from RGB-D data on Open3D.

Figure 7

Workflow of data acquisition and comparison of point clouds two by two based on absolute distances C2C.

Figure 8

Evolution of standard deviations [in mm] of measurements according to the number of averaged frames.

Figure 9

Evolution of the standard deviation of a pixel as a function of the number of averaged frames.

Figure 10

Comparison of the point cloud extracted from the Kinect Azure to the cloud from the TLS: (a) the differences between the two clouds [in m], (b) histogram of the values [in m].

Figure 11

Schema of the difference between the distance measured by the Kinect and the true horizontal distance from a point on the ground.

Figure 12

Evolution of the deviations, in mm, from the “true” distances.

Figure 13

Evolution of standard deviations of measurements as a function of distance [in mm].

Figure 14

Result of the indoor 3D reconstruction of the room: (a) global view of the room from the top; (b–d) focus on corners and objects.

Figure 15

Results of indoor 3D reconstruction of the office scene. (a) by keeping the windows uncovered; (b) by covering the windows.

Figure 16

Odometry drift effect highlighted by superposing point clouds from TLS (green) and point clouds from Kinect Azure (purple). (a) Top view of the point clouds, indicating the start point and the trajectory. (b) Perspective view. (c) Zoom on a corner of the room to highlight the odometry drift.

Figure 17

Means and standard deviations of C2C distances between the point clouds: (a) TLS vs. MLS, (b) TLS vs. Kinect Azure, (c) MLS vs. Kinect Azure.

Figure 18

Comparison of the intersection of the wall with the ground. (a) TLS (white) vs. Kinect Azure (color), (b) MLS (white) vs. Kinect Azure (color).

Figure 19

Multipath phenomenon observed on the intersection of two walls at the upper section corner of a room, as indicated by the red arrows. Pixels in black have zero depth values: depth image (left), passive infrared image (right). Microsoft.